TY - GEN
T1 - Coupled K-Nearest Centroid classification for non-iid data
AU - Li, Mu
AU - Li, Jinjiu
AU - Ou, Yuming
AU - Zhang, Ya
AU - Luo, Dan
AU - Bahtia, Maninder
AU - Cao, Longbing
N1 - This book: "is a special issue dedicated to the International Conference on Practical Applications on Agents and Multi-Agent Systems (PAAMS 2012 and PAAMS 2013) held in Salamanca during March 28–30, 2012 and May 22–24, 2013."
PY - 2014
Y1 - 2014
N2 - Most traditional classification methods assume the independence and identical distribution (iid) of objects, attributes and values. However, real world data, such as multi-agent data and behavioral data, usually contains strong couplings among values, attributes and objects, which greatly challenges existing methods and tools. This work targets the coupling similarities from these three perspectives and designs a novel classification method that applies a weighted K-Nearest Centroid to obtain the coupled similarity for non-iid data. From value and attribute perspectives, coupled similarity serves as a metric for nominal objects, which consider not only intra-coupled similarity within an attribute but also inter-coupled similarity between attributes. From the object perspective, we propose a more effective method that measures the centroid object by connecting all related objects. Extensive experiments on UCI and student data sets reveal that the proposed method outperforms classical methods for higher accuracy, especially in imbalanced data.
AB - Most traditional classification methods assume the independence and identical distribution (iid) of objects, attributes and values. However, real world data, such as multi-agent data and behavioral data, usually contains strong couplings among values, attributes and objects, which greatly challenges existing methods and tools. This work targets the coupling similarities from these three perspectives and designs a novel classification method that applies a weighted K-Nearest Centroid to obtain the coupled similarity for non-iid data. From value and attribute perspectives, coupled similarity serves as a metric for nominal objects, which consider not only intra-coupled similarity within an attribute but also inter-coupled similarity between attributes. From the object perspective, we propose a more effective method that measures the centroid object by connecting all related objects. Extensive experiments on UCI and student data sets reveal that the proposed method outperforms classical methods for higher accuracy, especially in imbalanced data.
UR - https://app.dimensions.ai/details/publication/pub.1089700130
UR - http://purl.org/au-research/grants/arc/DP1096218
UR - http://purl.org/au-research/grants/arc/DP0988016
UR - http://purl.org/au-research/grants/arc/LP100200774
UR - http://purl.org/au-research/grants/arc/LP0989721
UR - http://purl.org/au-research/grants/arc/LP100200774
U2 - 10.1007/978-3-662-45910-2_5
DO - 10.1007/978-3-662-45910-2_5
M3 - Conference proceeding contribution
SN - 9783662447499
T3 - Lecture Notes in Computer Science
SP - 89
EP - 100
BT - Transactions on computational collective intelligence XV
A2 - Nguyen, Ngoc Thanh
A2 - Kowalczyk, Ryszard
A2 - Corchado, Juan Manuel
A2 - Bajo, Javier
PB - Springer, Springer Nature
CY - Berlin
T2 - International Conference on Practical Applications on Agents and Multi-Agent Systems (2012)
Y2 - 28 March 2012 through 30 March 2012
ER -